By Sagar Shankaran, Founder of CallSphere
First-review permit rejections now track how readable your fee schedule is to AI assistants. The five pages a US building department should publish as text.
Key takeaways
Ask your Building Official that question and watch the face. The honest answer, in July 2026, is that a lot of them have not. The deck builder who pulls twenty permits a year in your town did not read Chapter 155. He asked an assistant on his phone, in his truck, at 6:40 in the morning, what he needs to submit for a 12-by-16 attached deck in your jurisdiction. Whatever that assistant told him is what shows up at your counter.
Which means your first-review rejection rate is no longer a measure of contractor carelessness. It is a measure of how readable your published documents are to a machine. If your master fee schedule is a scanned PDF adopted in 2019 and never re-posted, if your submittal checklist lives inside a portal that requires a login, if your code of ordinances is behind a viewer that does not hand out plain text, then the assistant guessed. It guessed using a neighboring town's rules, or a national code default, and the applicant is going to be annoyed at your permit technician about it.
This is not a future problem. It is what happened to the spring deck-and-pool rush this year.
Every permit technician has the drawer. The applications that came in missing a plot plan, or with the wrong fee attached, or from a contractor who is not registered with the town and does not know that registration is a separate form with a certificate of insurance. Each one is a phone call, a note in Accela or Tyler EnerGov or OpenGov, a resubmittal, and a wait.
The costs stack in three places. The counter, where 8:00 to 9:30 a.m. is already the worst window because contractors stop in on their way to job sites. The plan reviewer, who opens a file, finds it short, and closes it — and that opening and closing is not free. And the applicant's calendar, where a kickback that adds six days in late April is six days of a crew that was scheduled.
Being machine-readable, for a municipality, means the fee, the checklist, the code edition and the turnaround time are published as plain text on a public page with a date on it, so that an assistant answering a resident's question gets your actual rules instead of a guess. That is the entire ask. It is a records-publishing problem, not a software purchase.
Hear it before you finish reading
Talk to a live CallSphere AI voice agent in your browser — 60 seconds, no signup.
What changed in 2026 is that agent-to-agent protocols and machine-readable business surfaces went from a curiosity to a distribution channel. The party researching you may itself be software, working on someone's behalf: comparing, checking, shortlisting, sometimes booking. For a city that shows up in three ways at once, and all three matter to different people on your org chart.
A homeowner asks an assistant what she needs for a fence and whether she needs to go to the Zoning Board. A contractor's agent checks your setbacks, your impervious coverage limit, and your fee before he quotes the job — and if your fee is wrong by $180 he eats it or the homeowner argues about it. And a site selector working for a company evaluating three towns for a 40,000-square-foot building asks all three cities the same questions about plan review turnaround and pre-application process, and gets a comparison table without ever calling your Economic Development contact.
The unpleasant part is that being wrong is worse than being absent. If your site publishes nothing, the assistant says it does not know and the applicant calls you. If your site publishes a 2019 fee schedule as an image, the assistant reads whatever it can find and states it with confidence.
flowchart TD
A["Homeowner asks an assistant about a deck permit"] --> D
B["Contractor's agent checks setbacks and fees"] --> D
C["Site selector compares plan review turnaround"] --> D
D["What a machine can actually read on your site"]
D --> E["Current checklist, current fee, adopted code edition"]
D --> F["Missing, scanned, or out of date"]
E --> G["Complete application accepted at intake"]
F --> H["Kicked back, phone call to the permit tech"]
You do not need a project. You need five things published as ordinary web pages, in text, each with the date it took effect and the resolution or ordinance number that adopted it.
Add a sixth if you have boards: the Planning Board and Zoning Board of Appeals meeting calendar with the submittal deadline for each meeting, stated as a date, not as "21 days prior." An agent handles "September 3" reliably. It handles "21 days prior to the regular meeting" badly, and so do most applicants.
Illustration with stated assumptions. Suppose your department issues 780 residential permits a year, and 41% are rejected or held at first review for something that is not a design problem — missing document, wrong fee, unregistered contractor. That is 320 kickbacks. Assume each one costs 22 minutes of permit technician time to process and log, plus a 9-minute phone call.
| Measure | Today | After the five documents are fixed |
|---|---|---|
| Residential permits per year | 780 | 780 |
| First-review rejection rate | 41% | 26% (assumed) |
| Kickbacks per year | 320 | 203 |
| Staff time at 31 min each | 165 hours | 105 hours |
| Cost at $30/hr loaded | $4,950 | $3,150 |
| Average delay per kicked-back permit | 6 days | 6 days |
The $1,800 is not the reason to do this. Sixty staff hours back is roughly a month and a half of the 8:00 a.m. counter queue, and 117 projects that did not lose six days in the middle of construction season. The measure worth putting on the Manager's monthly report is the rejection rate itself, because it is the only number in a building department that tells you whether the outside world understands your rules. Pull it out of your permitting system as a baseline before you change anything, so you can show the difference in the fall.
Still reading? Stop comparing — try CallSphere live.
CallSphere ships complete AI voice agents per industry — 14 tools for healthcare, 10 agents for real estate, 4 specialists for salons. See how it actually handles a call before you book a demo.
Publishing clean documents will not make a bad application good. A deck built to the wrong span table is still going to fail inspection, and no amount of readable text prevents a contractor from ignoring what he was told. Expect the rejection rate to plateau at whatever your genuine design-error rate is.
It also will not settle interpretation. Half the questions your Building Official answers are judgment calls: whether an existing nonconforming setback carries, whether a structure counts toward impervious coverage, whether that is a repair or a replacement. Do not publish a page that pretends those have fixed answers, and do not let anyone route them to an assistant. Publish the process for getting a determination instead, with a name and a phone number attached.
And be careful about what publishing does to your obligations. Your fee schedule is adopted by the council; a web page that says something different from the adopted resolution is a problem, not a convenience. Put one person — usually the Deputy Clerk who processes the adopting resolution — in charge of updating the posted page the same week the council votes, and make it a step in the resolution routing, not a favor someone remembers.
No. Five pages of plain text on the website you already have, kept current, does the majority of the work. If your permitting system offers a public-facing page for fees and checklists, use it and make sure it is reachable without a login. The most common failure is not missing software, it is a PDF nobody re-posted after the June resolution.
It overlaps, but the standard is stricter. Search rewarded a page that ranked. What matters now is whether the answer extracted from your page is correct, current, and attributable to an adopted document. A page that ranks first and quotes a 2019 fee is worse for your department than no page at all.
Keep it — it is the authoritative copy and your attorney will insist. But do not rely on it as the only public statement of the handful of rules people actually ask about. Publish a short plain page for the common questions with a link back to the codified section, so the readable summary and the legal text agree and both are findable.
Ask an assistant on your own phone what is required for a deck permit in your jurisdiction, then hand the answer to your Building Official and let him mark it up. Ten minutes, no budget. Most departments find at least two wrong statements, and one of them is usually the fee.
The other half of this is the phone that rings when the published answer is not enough. CallSphere builds AI voice and chat agents that answer inbound lines and website chat, handle routine questions like fees, hours and inspection cutoffs, and take the caller's details when a permit technician needs to follow up. It works best on top of documents that are already correct, which is why the five pages come first.

Written by
Sagar Shankaran· Founder, CallSphere
LinkedInSagar Shankaran is the founder of CallSphere, where he builds production AI voice and chat agents deployed across healthcare, hospitality, real estate, and home services. He writes about agentic AI, LLM engineering, and shipping voice agents that handle real calls in production.
See how AI voice agents work for your industry. Live demo available -- no signup required.
How a US city sets per-department AI spend caps, alerts at 75% and 90%, and a single budget line — using the 2 July 2026 Claude Enterprise governance update.
A 6,100-page open-records request takes 23 days serially. Splitting the first pass across parallel agents cuts it to 6, with every redaction signed by a clerk.
The benefit cliff fills December and empties February. How 2026 forecasting handles dentistry's messy recall history, and the fee review that pays for it.
Cloned voices target municipal accounts payable. The callback, letterhead, dual-approval and prenote procedure that protects a city's vendor banking changes.
Supplement brands get adverse events reported by ticket, not by form. On-premises AI reads all 3,400 a month without that text ever leaving the building.
How a buyer's AI assistant screens consulting firms in 2026, why gated PDFs make you invisible, and the nine facts to publish in plain text this month.
© 2026 CallSphere Inc. All rights reserved.
Made within San Francisco
Watch how CallSphere handles real customer calls, schedules appointments, and processes payments — live.
Try Live DemoBook a DemoCalculate Your ROI